Re: Performance while loading data and indexing
Shridhar Daithankar <shridhar_daithankar@persistent.co.in>
From: "Shridhar Daithankar" <shridhar_daithankar@persistent.co.in>
To: pgsql-hackers@postgresql.org, pgsql-general@postgresql.org
Date: 2002-09-26T09:35:40Z
Lists: pgsql-hackers, pgsql-performance, pgsql-general
On 26 Sep 2002 at 19:17, Justin Clift wrote: > Shridhar Daithankar wrote: > <snip> > > 3) Any suggsestions for runtime as data load and query will be going in > > parallel. > > That sounds unusual. From reading this, it *sounds* like you'll be > running queries against an incomplete dataset, or maybe just running the > queries that affect the tables loaded thus far (during the initial > load). That's correct. Load the data so far and keep inserting data as and when it generates. They don't mind running against data so far. It's not very accurate stuff IMO... > > fsync=true (Sad but true. Left untouched.. Will that make a difference on > > SCSI?) > > Definitely. Have directly measured a ~ 2x tps throughput increase on > FreeBSD when leaving fsync off whilst performance measuring stuff > recently (PG 7.2.2). Like anything it'll depend on workload, phase of > moon, etc, but it's a decent indicator. I didn't know even that matters with SCSI..Will check out.. > fsync = off would help during the data load, but not a good idea if > you're going to be running queries against it at the same time. That's OK for the reasons mentioned above. It wouldn't be out of place to expect a UPS to such an installation... Bye Shridhar -- Hoare's Law of Large Problems: Inside every large problem is a small problem struggling to get out.